Artist Prompt: How to Write an Exact Assignment for Image Generation

Why the first result requires refinement

The first run of the tool almost always gives a general direction, but not an exact image. This is normal: you have not yet refined composition, light and palette. For example, if you enter the idea 'quiet bookstore on a rainy evening', the tool may suggest a scene with warm light in the windows, but without indicating where the camera is, what is in the foreground and what details to exclude. Such a prompt is already useful as a starting point, but it needs refinement.

Important: the tool does not analyze images and has no style presets. It works only with your text. Therefore the more specifically you describe the desired, the more precise the result. Do not be afraid to rewrite the 'Your text or idea' field several times, adding new details. Each iteration brings you closer to the goal.

First prompt: 'Bookstore on a rainy evening, warm light in the windows'. Result: general plan, but no detail focus.

How to Refine Composition

Composition is the arrangement of objects in the frame. Add instructions to the prompt: 'first-person view', 'close-up of the window display', 'wide angle, store on left, street on right'. This will help the generative model or artist understand what exactly should be the center of attention. If you want the viewer to feel inside the scene, specify 'camera at eye level, gaze directed at the store door'.

Do not forget about the foreground and background. For example: 'on the foreground wet cobblestone with reflections, on the background blurred silhouettes of pedestrians'. Such details create depth. If the background was empty in the first result, add 'light fog, trees in the distance'. The more specificity, the better.

Refined prompt: 'First-person view, bookstore on the left, wet cobblestone in the foreground, blurred pedestrians in the background, rainy evening'.

Lighting and Palette

Lighting sets the mood. In the 'Tone, style or format' field you can specify: 'warm amber light from windows', 'cold blue twilight', 'contrast between warm and cold'. This will help avoid a flat image. If the first result was too dark, add 'soft diffused light, shadows not too deep'.

Palette is a set of colors. Describe it in words: 'muted green and brown tones', 'bright red accents on a gray background'. You do not need to list all colors, just set the general mood. For example, 'palette: ochre, dark blue, warm yellow'. This will help make the image coherent.

Prompt with lighting: 'Warm light from the window, cold blue evening air, contrast, palette: ochre, dark blue, yellow'.

Exclusions: What Should Not Get into the Frame

Exclusions are as important as inclusions. If you do not want to see people, cars or bright signs in the image, specify this in the prompt. For example: 'no people, no cars, no neon signs'. This will help avoid unnecessary details that distract from the main subject.

Sometimes the first result contains unwanted elements: for example, too much rain or glare. Add 'moderate rain, no strong glares'. If you want minimalism, write 'no extra items, only the window display and cobblestone'. Clear exclusions save time on subsequent edits.

Prompt with exclusions: '...no people, no cars, no neon signs, moderate rain'.

Practical Iteration: from Draft to Finish

Start with a simple prompt, get the result, then analyze what is wrong. For example, if the shop is too far in the picture, add 'close-up of the window display'. If the light is too cold, replace 'cold blue' with 'warm amber'. Each edit should be specific. Do not be afraid to rewrite the whole text if needed.

The tool allows entering up to 8000 characters in the first field and up to 1000 characters in the details field. This is enough to describe a complex scene. But do not overload the prompt: if there are too many details, the model may get confused. Better to make several iterations, gradually adding new elements.

Remember, the tool does not guarantee an ideal result from the first time. It is only an assistant in formulation. You can download the obtained prompt in TXT and use it in any image generator or give it to an artist.

Iteration 1: 'Bookstore, rain, evening'. Iteration 2: 'First-person view, warm light, wet cobblestone, no people'. Iteration 3: '...palette: ochre and blue, exclude machines'.

How to Check the Result

After generating the prompt read it aloud. Is it clear what should be in the image? Are there any contradictions? For example, 'warm light' and 'cold blue' can conflict if you do not specify that it is a contrast. Make sure all important details are included, and unwanted ones are excluded.

If you use the prompt for a generative model, check whether the result matches your description. If not, return to the tool and refine the text. Do not be afraid to experiment with formulations. Sometimes replacing one word changes everything.

Also pay attention to the output format. The tool outputs plain text that can be copied or downloaded. Make sure you saved it in a convenient location.

Check: 'The prompt says 'no people', but there are 'silhouettes of pedestrians' — this is a contradiction. Need to remove one of the two'.

Tool Limitations

The tool does not generate images, has no style presets, sliders and levels of changes. It does not save history, does not perform batch processing and does not integrate with other services. You cannot upload documents or use OCR. Everything it does is create a textual prompt based on your input.

Maximum length of the first field is 8000 characters, details field — 1000 characters. For images the limit is 4 MB, for audio — 20 MB or 5 minutes. But remember: the tool does not work with audio or video, only with text. Model responses may contain errors, and queue and availability may change.

Do not expect the tool to write a ready-made masterpiece for you. It is only a draft that needs refinement. Use it as a starting point for your own creativity.

Limitation: cannot upload a reference image. All description must be textual.